Risk parity: how it works, how to size it, and a backtest

Risk parity is an allocation rule that sizes positions by risk rather than by capital. Instead of putting 60 percent of the money in stocks and 40 percent in bonds, it asks how much each asset moves and weights them so that each contributes an equal share of the portfolio's volatility. Because stocks are typically two to three times as volatile as Treasury bonds, a risk parity portfolio ends up holding far fewer dollars of stocks than a traditional balanced portfolio. The simplest implementation, inverse-volatility weighting, can be written in one sentence and is the prompt at the bottom of this page.

What problem is risk parity solving?

A 60/40 portfolio looks balanced by dollars and is not balanced by risk. Because equity volatility is so much higher than bond volatility, roughly 90 percent of a 60/40 portfolio's variance has historically come from the equity sleeve. When stocks fall, the portfolio falls; the bonds soften the blow but do not change the shape of the outcome. Diversification in dollars is not diversification in risk.

Risk parity flips the question. Rather than deciding how many dollars go to each asset, it decides how much risk each asset should contribute and backs out the dollars. Edward Qian's 2005 paper Risk Parity Portfolios formalized the idea and coined the term. Bridgewater's All Weather fund, launched in 1996, applied a version of it years earlier, motivated by the observation that different assets do well in different combinations of growth and inflation, and that a portfolio balanced across those environments should hold up in more of them.

What are the exact rules?

The full risk parity solution equalizes each asset's contribution to portfolio variance, which requires the covariance matrix and an iterative solver. The version tested here is the simpler inverse-volatility approximation, which ignores correlations and is what most individuals mean when they say risk parity:

1. On the last trading day of each month, compute each asset's realized volatility over the trailing 60 trading days.
2. Set each asset's raw weight to one divided by its volatility.
3. Normalize the weights so they sum to 100 percent.
4. Hold those weights until the next month end.

With three assets, if SPY's volatility is 16 percent, TLT's is 14 percent, and GLD's is 15 percent, the weights are close to equal. If SPY's volatility spikes to 32 percent in a selloff, its weight roughly halves, and the portfolio automatically de-risks. That automatic response is the second thing risk parity does, beyond static balance: it trims whatever has recently become dangerous.

The three-asset universe of US stocks, long Treasuries, and gold is the common starting point because the three have historically responded to different conditions. Stocks want growth, long bonds want falling rates and disinflation, gold wants inflation or crisis. No universe guarantees that at least one asset rises, but this one has spread its bets across the main regimes.

Why might it work?

Two mechanisms are at work, and it helps to separate them.

The first is balance. If asset returns are roughly uncorrelated and you have no view on which will do best, equal risk contribution is close to the allocation that maximizes diversification. Asness, Frazzini, and Pedersen's 2012 paper Leverage Aversion and Risk Parity argues that low-volatility assets like bonds have historically offered higher risk-adjusted returns than high-volatility assets like stocks, because most investors cannot or will not lever, and so they overpay for the volatile assets that deliver returns without leverage. If that is true, overweighting bonds by risk is compensated.

The second is volatility responsiveness. Volatility clusters: high-volatility days are followed by more high-volatility days. Because inverse-volatility weights shrink an asset's position when its volatility rises, the portfolio tends to reduce exposure early in a selloff and add it back as conditions calm. This does not predict direction, but it has historically smoothed the return path relative to fixed weights.

When does it fail?

Risk parity depends on the assets not falling together. When they do, it has no defense.

Correlated selloffs. 2022 is the recent example. Rising rates pushed stocks and long Treasuries down at the same time, and gold was flat. A three-asset inverse-volatility portfolio had nowhere to hide, and because it held more bonds than a 60/40 portfolio, the bond drawdown hurt more. Any period in which inflation surprises to the upside is dangerous for a portfolio that is structurally long duration.

Rising rate regimes generally. The strategy's historical record was built mostly during a forty-year decline in interest rates, which made long Treasuries both a diversifier and a return engine. In a rising-rate regime, the largest position in the portfolio has a headwind. Whether the diversification benefit survives without the return is an open question that the historical sample cannot fully answer. See the common mistakes guide on regime-limited samples.

Volatility whipsaws. Because weights respond to trailing volatility, a sharp spike followed by a fast recovery can cause the portfolio to cut an asset after the damage and miss the rebound. The mechanism that helps in a prolonged bear market hurts in a V-shaped one.

Unlevered expected returns. Without leverage, an inverse-volatility portfolio is bond-heavy and has a lower expected return than an equity-heavy one. Its appeal is risk-adjusted, not absolute. An investor who benchmarks against the S&P 500 will spend most bull markets behind it.

What does a backtest show?

The card below is a live tear sheet of an ENSEMBLE model that applies inverse-volatility weighting across gold, long Treasuries, and broad commodities on a 63-day lookback. It is the same weighting rule as the prompt on this page applied to a defensive, equity-free universe, which is how the approach is often used as a sleeve alongside an equity core. The figures are computed by the platform from daily closes, net of a simulated transaction cost, and refresh nightly.

On the full tear sheet, look at the allocation chart over time. The weights are never static; they shift month to month as each asset's recent volatility changes. Compare the weight of each asset in calm periods and in stressed ones to see the de-risking mechanism at work. Then check the drawdown chart against the drawdown of any single component to see what the balance bought.

To build the three-asset stock, bond, and gold version, run the prompt below. To see the metrics on the card defined, see backtest metrics explained.

How to run it and change it

Variations worth testing:

• Lookback. Change 60 days to 20 or 120. Shorter windows make the weights more reactive and the portfolio more expensive to run.
• Universe. Add DBC for broad commodities, or replace TLT with IEF for intermediate rather than long duration. Intermediate bonds reduce the rate sensitivity that hurt in 2022 and also reduce the diversification benefit in equity selloffs.
• Volatility target. Ask for the portfolio to be scaled so its trailing volatility targets 10 percent, holding the remainder in BIL. This adds a cash buffer in stressed periods without leverage.
• Rebalance cadence. Quarterly rebalancing cuts turnover and makes the weights lag further behind volatility changes.

Practical notes

Inverse-volatility weighting is not full risk parity, because it ignores correlations. When two assets are highly correlated, inverse-volatility overweights their combined risk. For a three-asset universe with historically low correlations, the approximation is usually close; for a universe with several equity funds, it is not, and an equal-risk-contribution optimizer is the right tool. ENSEMBLE can build either; describe the one you mean.

The strategy is rate sensitive by construction. Whatever the volatility of each asset, the largest dollar weight will usually be in the bond fund. Investors who want the balancing logic without that much duration often shorten the bond position or add a trend filter to it, along the lines of the 200-day moving average rule.

Finally, the portfolio is fully invested at all times in this version. It never moves to cash. That is a feature in the sense that it does not try to time markets, and a limitation in the sense that in a correlated selloff its only response is to shift weights among falling assets. Pairing it with a trend-following strategy such as dual momentum in an ENSEMBLE portfolio is one way to add a cash exit without changing the risk parity rule itself.

Backtests are hypothetical, past performance does not guarantee future results, and ENSEMBLE is research software rather than an investment adviser.